s390/kernel: introduce fpu-internal.h with fpu helper functions

Introduce a new structure to manage FP and VX registers. Refactor the
save and restore of floating point and vector registers with a set
of helper functions in fpu-internal.h.

Signed-off-by: Hendrik Brueckner <brueckner@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>

9 files changed, 299 insertions, 285 deletions

diff --git a/arch/s390/include/asm/fpu-internal.h b/arch/s390/include/asm/fpu-internal.h
new file mode 100644
index 0000000..04b4cfc
--- /dev/null
+++ b/arch/s390/include/asm/fpu-internal.h
@@ -0,0 +1,191 @@
+/*
+ * General floating pointer and vector register helpers
+ *
+ * Copyright IBM Corp. 2015
+ * Author(s): Hendrik Brueckner <brueckner@linux.vnet.ibm.com>
+ */
+
+#ifndef _ASM_S390_FPU_INTERNAL_H
+#define _ASM_S390_FPU_INTERNAL_H
+
+#include <linux/errno.h>
+#include <linux/string.h>
+#include <asm/linkage.h>
+#include <asm/ctl_reg.h>
+#include <asm/sigcontext.h>
+
+struct fpu {
+	__u32 fpc;			/* Floating-point control */
+	__u32 pad;
+	freg_t fprs[__NUM_FPRS];	/* Floating-point register save area */
+	__vector128 *vxrs;		/* Vector register save area */
+};
+
+#define is_vx_fpu(fpu) (!!(fpu)->vxrs)
+#define is_vx_task(tsk) (!!(tsk)->thread.fpu.vxrs)
+
+static inline int test_fp_ctl(u32 fpc)
+{
+	u32 orig_fpc;
+	int rc;
+
+	asm volatile(
+		"	efpc    %1\n"
+		"	sfpc	%2\n"
+		"0:	sfpc	%1\n"
+		"	la	%0,0\n"
+		"1:\n"
+		EX_TABLE(0b,1b)
+		: "=d" (rc), "=d" (orig_fpc)
+		: "d" (fpc), "0" (-EINVAL));
+	return rc;
+}
+
+static inline void save_fp_ctl(u32 *fpc)
+{
+	asm volatile(
+		"       stfpc   %0\n"
+		: "+Q" (*fpc));
+}
+
+static inline int restore_fp_ctl(u32 *fpc)
+{
+	int rc;
+
+	asm volatile(
+		"	lfpc    %1\n"
+		"0:	la	%0,0\n"
+		"1:\n"
+		: "=d" (rc) : "Q" (*fpc), "0" (-EINVAL));
+	return rc;
+}
+
+static inline void save_fp_regs(freg_t *fprs)
+{
+	asm volatile("std 0,%0" : "=Q" (fprs[0]));
+	asm volatile("std 2,%0" : "=Q" (fprs[2]));
+	asm volatile("std 4,%0" : "=Q" (fprs[4]));
+	asm volatile("std 6,%0" : "=Q" (fprs[6]));
+	asm volatile("std 1,%0" : "=Q" (fprs[1]));
+	asm volatile("std 3,%0" : "=Q" (fprs[3]));
+	asm volatile("std 5,%0" : "=Q" (fprs[5]));
+	asm volatile("std 7,%0" : "=Q" (fprs[7]));
+	asm volatile("std 8,%0" : "=Q" (fprs[8]));
+	asm volatile("std 9,%0" : "=Q" (fprs[9]));
+	asm volatile("std 10,%0" : "=Q" (fprs[10]));
+	asm volatile("std 11,%0" : "=Q" (fprs[11]));
+	asm volatile("std 12,%0" : "=Q" (fprs[12]));
+	asm volatile("std 13,%0" : "=Q" (fprs[13]));
+	asm volatile("std 14,%0" : "=Q" (fprs[14]));
+	asm volatile("std 15,%0" : "=Q" (fprs[15]));
+}
+
+static inline void restore_fp_regs(freg_t *fprs)
+{
+	asm volatile("ld 0,%0" : : "Q" (fprs[0]));
+	asm volatile("ld 2,%0" : : "Q" (fprs[2]));
+	asm volatile("ld 4,%0" : : "Q" (fprs[4]));
+	asm volatile("ld 6,%0" : : "Q" (fprs[6]));
+	asm volatile("ld 1,%0" : : "Q" (fprs[1]));
+	asm volatile("ld 3,%0" : : "Q" (fprs[3]));
+	asm volatile("ld 5,%0" : : "Q" (fprs[5]));
+	asm volatile("ld 7,%0" : : "Q" (fprs[7]));
+	asm volatile("ld 8,%0" : : "Q" (fprs[8]));
+	asm volatile("ld 9,%0" : : "Q" (fprs[9]));
+	asm volatile("ld 10,%0" : : "Q" (fprs[10]));
+	asm volatile("ld 11,%0" : : "Q" (fprs[11]));
+	asm volatile("ld 12,%0" : : "Q" (fprs[12]));
+	asm volatile("ld 13,%0" : : "Q" (fprs[13]));
+	asm volatile("ld 14,%0" : : "Q" (fprs[14]));
+	asm volatile("ld 15,%0" : : "Q" (fprs[15]));
+}
+
+static inline void save_vx_regs(__vector128 *vxrs)
+{
+	typedef struct { __vector128 _[__NUM_VXRS]; } addrtype;
+
+	asm volatile(
+		"	la	1,%0\n"
+		"	.word	0xe70f,0x1000,0x003e\n"	/* vstm 0,15,0(1) */
+		"	.word	0xe70f,0x1100,0x0c3e\n"	/* vstm 16,31,256(1) */
+		: "=Q" (*(addrtype *) vxrs) : : "1");
+}
+
+static inline void save_vx_regs_safe(__vector128 *vxrs)
+{
+	unsigned long cr0, flags;
+
+	flags = arch_local_irq_save();
+	__ctl_store(cr0, 0, 0);
+	__ctl_set_bit(0, 17);
+	__ctl_set_bit(0, 18);
+	save_vx_regs(vxrs);
+	__ctl_load(cr0, 0, 0);
+	arch_local_irq_restore(flags);
+}
+
+static inline void restore_vx_regs(__vector128 *vxrs)
+{
+	typedef struct { __vector128 _[__NUM_VXRS]; } addrtype;
+
+	asm volatile(
+		"	la	1,%0\n"
+		"	.word	0xe70f,0x1000,0x0036\n"	/* vlm 0,15,0(1) */
+		"	.word	0xe70f,0x1100,0x0c36\n"	/* vlm 16,31,256(1) */
+		: : "Q" (*(addrtype *) vxrs) : "1");
+}
+
+static inline void convert_vx_to_fp(freg_t *fprs, __vector128 *vxrs)
+{
+	int i;
+
+	for (i = 0; i < __NUM_FPRS; i++)
+		fprs[i] = *(freg_t *)(vxrs + i);
+}
+
+static inline void convert_fp_to_vx(__vector128 *vxrs, freg_t *fprs)
+{
+	int i;
+
+	for (i = 0; i < __NUM_FPRS; i++)
+		*(freg_t *)(vxrs + i) = fprs[i];
+}
+
+static inline void fpregs_store(_s390_fp_regs *fpregs, struct fpu *fpu)
+{
+	fpregs->pad = 0;
+	if (is_vx_fpu(fpu))
+		convert_vx_to_fp((freg_t *)&fpregs->fprs, fpu->vxrs);
+	else
+		memcpy((freg_t *)&fpregs->fprs, fpu->fprs,
+		       sizeof(fpregs->fprs));
+}
+
+static inline void fpregs_load(_s390_fp_regs *fpregs, struct fpu *fpu)
+{
+	if (is_vx_fpu(fpu))
+		convert_fp_to_vx(fpu->vxrs, (freg_t *)&fpregs->fprs);
+	else
+		memcpy(fpu->fprs, (freg_t *)&fpregs->fprs,
+		       sizeof(fpregs->fprs));
+}
+
+static inline void save_fpu_regs(struct fpu *fpu)
+{
+	save_fp_ctl(&fpu->fpc);
+	if (is_vx_fpu(fpu))
+		save_vx_regs(fpu->vxrs);
+	else
+		save_fp_regs(fpu->fprs);
+}
+
+static inline void restore_fpu_regs(struct fpu *fpu)
+{
+	restore_fp_ctl(&fpu->fpc);
+	if (is_vx_fpu(fpu))
+		restore_vx_regs(fpu->vxrs);
+	else
+		restore_fp_regs(fpu->fprs);
+}
+
+#endif /* _ASM_S390_FPU_INTERNAL_H */
diff --git a/arch/s390/include/asm/processor.h b/arch/s390/include/asm/processor.h
index f4d9f74..19f51db 100644
--- a/arch/s390/include/asm/processor.h
+++ b/arch/s390/include/asm/processor.h
@@ -28,6 +28,7 @@
 #include <asm/ptrace.h>
 #include <asm/setup.h>
 #include <asm/runtime_instr.h>
+#include <asm/fpu-internal.h>
 
 static inline void set_cpu_flag(int flag)
 {
@@ -85,7 +86,7 @@ typedef struct {
  * Thread structure
  */
 struct thread_struct {
-	s390_fp_regs fp_regs;
+	struct fpu fpu;			/* FP and VX register save area */
 	unsigned int  acrs[NUM_ACRS];
         unsigned long ksp;              /* kernel stack pointer             */
 	mm_segment_t mm_segment;
@@ -101,7 +102,6 @@ struct thread_struct {
 	struct runtime_instr_cb *ri_cb;
 	int ri_signum;
 	unsigned char trap_tdb[256];	/* Transaction abort diagnose block */
-	__vector128 *vxrs;		/* Vector register save area */
 };
 
 /* Flag to disable transactions. */
diff --git a/arch/s390/include/asm/switch_to.h b/arch/s390/include/asm/switch_to.h
index b304031..caf4f23 100644
--- a/arch/s390/include/asm/switch_to.h
+++ b/arch/s390/include/asm/switch_to.h
@@ -8,138 +8,12 @@
 #define __ASM_SWITCH_TO_H
 
 #include <linux/thread_info.h>
+#include <asm/fpu-internal.h>
 #include <asm/ptrace.h>
 
 extern struct task_struct *__switch_to(void *, void *);
 extern void update_cr_regs(struct task_struct *task);
 
-static inline int test_fp_ctl(u32 fpc)
-{
-	u32 orig_fpc;
-	int rc;
-
-	asm volatile(
-		"	efpc    %1\n"
-		"	sfpc	%2\n"
-		"0:	sfpc	%1\n"
-		"	la	%0,0\n"
-		"1:\n"
-		EX_TABLE(0b,1b)
-		: "=d" (rc), "=d" (orig_fpc)
-		: "d" (fpc), "0" (-EINVAL));
-	return rc;
-}
-
-static inline void save_fp_ctl(u32 *fpc)
-{
-	asm volatile(
-		"       stfpc   %0\n"
-		: "+Q" (*fpc));
-}
-
-static inline int restore_fp_ctl(u32 *fpc)
-{
-	int rc;
-
-	asm volatile(
-		"	lfpc    %1\n"
-		"0:	la	%0,0\n"
-		"1:\n"
-		: "=d" (rc) : "Q" (*fpc), "0" (-EINVAL));
-	return rc;
-}
-
-static inline void save_fp_regs(freg_t *fprs)
-{
-	asm volatile("std 0,%0" : "=Q" (fprs[0]));
-	asm volatile("std 2,%0" : "=Q" (fprs[2]));
-	asm volatile("std 4,%0" : "=Q" (fprs[4]));
-	asm volatile("std 6,%0" : "=Q" (fprs[6]));
-	asm volatile("std 1,%0" : "=Q" (fprs[1]));
-	asm volatile("std 3,%0" : "=Q" (fprs[3]));
-	asm volatile("std 5,%0" : "=Q" (fprs[5]));
-	asm volatile("std 7,%0" : "=Q" (fprs[7]));
-	asm volatile("std 8,%0" : "=Q" (fprs[8]));
-	asm volatile("std 9,%0" : "=Q" (fprs[9]));
-	asm volatile("std 10,%0" : "=Q" (fprs[10]));
-	asm volatile("std 11,%0" : "=Q" (fprs[11]));
-	asm volatile("std 12,%0" : "=Q" (fprs[12]));
-	asm volatile("std 13,%0" : "=Q" (fprs[13]));
-	asm volatile("std 14,%0" : "=Q" (fprs[14]));
-	asm volatile("std 15,%0" : "=Q" (fprs[15]));
-}
-
-static inline void restore_fp_regs(freg_t *fprs)
-{
-	asm volatile("ld 0,%0" : : "Q" (fprs[0]));
-	asm volatile("ld 2,%0" : : "Q" (fprs[2]));
-	asm volatile("ld 4,%0" : : "Q" (fprs[4]));
-	asm volatile("ld 6,%0" : : "Q" (fprs[6]));
-	asm volatile("ld 1,%0" : : "Q" (fprs[1]));
-	asm volatile("ld 3,%0" : : "Q" (fprs[3]));
-	asm volatile("ld 5,%0" : : "Q" (fprs[5]));
-	asm volatile("ld 7,%0" : : "Q" (fprs[7]));
-	asm volatile("ld 8,%0" : : "Q" (fprs[8]));
-	asm volatile("ld 9,%0" : : "Q" (fprs[9]));
-	asm volatile("ld 10,%0" : : "Q" (fprs[10]));
-	asm volatile("ld 11,%0" : : "Q" (fprs[11]));
-	asm volatile("ld 12,%0" : : "Q" (fprs[12]));
-	asm volatile("ld 13,%0" : : "Q" (fprs[13]));
-	asm volatile("ld 14,%0" : : "Q" (fprs[14]));
-	asm volatile("ld 15,%0" : : "Q" (fprs[15]));
-}
-
-static inline void save_vx_regs(__vector128 *vxrs)
-{
-	typedef struct { __vector128 _[__NUM_VXRS]; } addrtype;
-
-	asm volatile(
-		"	la	1,%0\n"
-		"	.word	0xe70f,0x1000,0x003e\n"	/* vstm 0,15,0(1) */
-		"	.word	0xe70f,0x1100,0x0c3e\n"	/* vstm 16,31,256(1) */
-		: "=Q" (*(addrtype *) vxrs) : : "1");
-}
-
-static inline void save_vx_regs_safe(__vector128 *vxrs)
-{
-	unsigned long cr0, flags;
-
-	flags = arch_local_irq_save();
-	__ctl_store(cr0, 0, 0);
-	__ctl_set_bit(0, 17);
-	__ctl_set_bit(0, 18);
-	save_vx_regs(vxrs);
-	__ctl_load(cr0, 0, 0);
-	arch_local_irq_restore(flags);
-}
-
-static inline void restore_vx_regs(__vector128 *vxrs)
-{
-	typedef struct { __vector128 _[__NUM_VXRS]; } addrtype;
-
-	asm volatile(
-		"	la	1,%0\n"
-		"	.word	0xe70f,0x1000,0x0036\n"	/* vlm 0,15,0(1) */
-		"	.word	0xe70f,0x1100,0x0c36\n"	/* vlm 16,31,256(1) */
-		: : "Q" (*(addrtype *) vxrs) : "1");
-}
-
-static inline void save_fp_vx_regs(struct task_struct *task)
-{
-	if (task->thread.vxrs)
-		save_vx_regs(task->thread.vxrs);
-	else
-		save_fp_regs(task->thread.fp_regs.fprs);
-}
-
-static inline void restore_fp_vx_regs(struct task_struct *task)
-{
-	if (task->thread.vxrs)
-		restore_vx_regs(task->thread.vxrs);
-	else
-		restore_fp_regs(task->thread.fp_regs.fprs);
-}
-
 static inline void save_access_regs(unsigned int *acrs)
 {
 	typedef struct { int _[NUM_ACRS]; } acrstype;
@@ -156,15 +30,13 @@ static inline void restore_access_regs(unsigned int *acrs)
 
 #define switch_to(prev,next,last) do {					\
 	if (prev->mm) {							\
-		save_fp_ctl(&prev->thread.fp_regs.fpc);			\
-		save_fp_vx_regs(prev);					\
+		save_fpu_regs(&prev->thread.fpu);			\
 		save_access_regs(&prev->thread.acrs[0]);		\
 		save_ri_cb(prev->thread.ri_cb);				\
 	}								\
 	if (next->mm) {							\
 		update_cr_regs(next);					\
-		restore_fp_ctl(&next->thread.fp_regs.fpc);		\
-		restore_fp_vx_regs(next);				\
+		restore_fpu_regs(&next->thread.fpu);			\
 		restore_access_regs(&next->thread.acrs[0]);		\
 		restore_ri_cb(next->thread.ri_cb, prev->thread.ri_cb);	\
 	}								\
diff --git a/arch/s390/kernel/compat_signal.c b/arch/s390/kernel/compat_signal.c
index b55d206..4529951 100644
--- a/arch/s390/kernel/compat_signal.c
+++ b/arch/s390/kernel/compat_signal.c
@@ -153,33 +153,15 @@ int copy_siginfo_from_user32(siginfo_t *to, compat_siginfo_t __user *from)
 /* Store registers needed to create the signal frame */
 static void store_sigregs(void)
 {
-	int i;
-
 	save_access_regs(current->thread.acrs);
-	save_fp_ctl(&current->thread.fp_regs.fpc);
-	if (current->thread.vxrs) {
-		save_vx_regs(current->thread.vxrs);
-		for (i = 0; i < __NUM_FPRS; i++)
-			current->thread.fp_regs.fprs[i] =
-				*(freg_t *)(current->thread.vxrs + i);
-	} else
-		save_fp_regs(current->thread.fp_regs.fprs);
+	save_fpu_regs(&current->thread.fpu);
 }
 
 /* Load registers after signal return */
 static void load_sigregs(void)
 {
-	int i;
-
 	restore_access_regs(current->thread.acrs);
-	restore_fp_ctl(&current->thread.fp_regs.fpc);
-	if (current->thread.vxrs) {
-		for (i = 0; i < __NUM_FPRS; i++)
-			*(freg_t *)(current->thread.vxrs + i) =
-				current->thread.fp_regs.fprs[i];
-		restore_vx_regs(current->thread.vxrs);
-	} else
-		restore_fp_regs(current->thread.fp_regs.fprs);
+	restore_fpu_regs(&current->thread.fpu);
 }
 
 static int save_sigregs32(struct pt_regs *regs, _sigregs32 __user *sregs)
@@ -196,8 +178,7 @@ static int save_sigregs32(struct pt_regs *regs, _sigregs32 __user *sregs)
 		user_sregs.regs.gprs[i] = (__u32) regs->gprs[i];
 	memcpy(&user_sregs.regs.acrs, current->thread.acrs,
 	       sizeof(user_sregs.regs.acrs));
-	memcpy(&user_sregs.fpregs, &current->thread.fp_regs,
-	       sizeof(user_sregs.fpregs));
+	fpregs_store((_s390_fp_regs *) &user_sregs.fpregs, &current->thread.fpu);
 	if (__copy_to_user(sregs, &user_sregs, sizeof(_sigregs32)))
 		return -EFAULT;
 	return 0;
@@ -235,9 +216,7 @@ static int restore_sigregs32(struct pt_regs *regs,_sigregs32 __user *sregs)
 		regs->gprs[i] = (__u64) user_sregs.regs.gprs[i];
 	memcpy(&current->thread.acrs, &user_sregs.regs.acrs,
 	       sizeof(current->thread.acrs));
-
-	memcpy(&current->thread.fp_regs, &user_sregs.fpregs,
-	       sizeof(current->thread.fp_regs));
+	fpregs_load((_s390_fp_regs *) &user_sregs.fpregs, &current->thread.fpu);
 
 	clear_pt_regs_flag(regs, PIF_SYSCALL); /* No longer in a system call */
 	return 0;
@@ -258,13 +237,13 @@ static int save_sigregs_ext32(struct pt_regs *regs,
 		return -EFAULT;
 
 	/* Save vector registers to signal stack */
-	if (current->thread.vxrs) {
+	if (is_vx_task(current)) {
 		for (i = 0; i < __NUM_VXRS_LOW; i++)
-			vxrs[i] = *((__u64 *)(current->thread.vxrs + i) + 1);
+			vxrs[i] = *((__u64 *)(current->thread.fpu.vxrs + i) + 1);
 		if (__copy_to_user(&sregs_ext->vxrs_low, vxrs,
 				   sizeof(sregs_ext->vxrs_low)) ||
 		    __copy_to_user(&sregs_ext->vxrs_high,
-				   current->thread.vxrs + __NUM_VXRS_LOW,
+				   current->thread.fpu.vxrs + __NUM_VXRS_LOW,
 				   sizeof(sregs_ext->vxrs_high)))
 			return -EFAULT;
 	}
@@ -286,15 +265,15 @@ static int restore_sigregs_ext32(struct pt_regs *regs,
 		*(__u32 *)&regs->gprs[i] = gprs_high[i];
 
 	/* Restore vector registers from signal stack */
-	if (current->thread.vxrs) {
+	if (is_vx_task(current)) {
 		if (__copy_from_user(vxrs, &sregs_ext->vxrs_low,
 				     sizeof(sregs_ext->vxrs_low)) ||
-		    __copy_from_user(current->thread.vxrs + __NUM_VXRS_LOW,
+		    __copy_from_user(current->thread.fpu.vxrs + __NUM_VXRS_LOW,
 				     &sregs_ext->vxrs_high,
 				     sizeof(sregs_ext->vxrs_high)))
 			return -EFAULT;
 		for (i = 0; i < __NUM_VXRS_LOW; i++)
-			*((__u64 *)(current->thread.vxrs + i) + 1) = vxrs[i];
+			*((__u64 *)(current->thread.fpu.vxrs + i) + 1) = vxrs[i];
 	}
 	return 0;
 }
@@ -472,7 +451,7 @@ static int setup_rt_frame32(struct ksignal *ksig, sigset_t *set,
 	 */
 	uc_flags = UC_GPRS_HIGH;
 	if (MACHINE_HAS_VX) {
-		if (current->thread.vxrs)
+		if (is_vx_task(current))
 			uc_flags |= UC_VXRS;
 	} else
 		frame_size -= sizeof(frame->uc.uc_mcontext_ext.vxrs_low) +
diff --git a/arch/s390/kernel/nmi.c b/arch/s390/kernel/nmi.c
index 56b5508..e66141c 100644
--- a/arch/s390/kernel/nmi.c
+++ b/arch/s390/kernel/nmi.c
@@ -21,6 +21,7 @@
 #include <asm/nmi.h>
 #include <asm/crw.h>
 #include <asm/switch_to.h>
+#include <asm/fpu-internal.h>
 #include <asm/ctl_reg.h>
 
 struct mcck_struct {
diff --git a/arch/s390/kernel/process.c b/arch/s390/kernel/process.c
index 8f587d8..61795bc 100644
--- a/arch/s390/kernel/process.c
+++ b/arch/s390/kernel/process.c
@@ -81,8 +81,8 @@ void release_thread(struct task_struct *dead_task)
 
 void arch_release_task_struct(struct task_struct *tsk)
 {
-	if (tsk->thread.vxrs)
-		kfree(tsk->thread.vxrs);
+	if (is_vx_task(tsk))
+		kfree(tsk->thread.fpu.vxrs);
 }
 
 int copy_thread(unsigned long clone_flags, unsigned long new_stackp,
@@ -143,10 +143,10 @@ int copy_thread(unsigned long clone_flags, unsigned long new_stackp,
 	frame->childregs.psw.mask &= ~PSW_MASK_RI;
 
 	/* Save the fpu registers to new thread structure. */
-	save_fp_ctl(&p->thread.fp_regs.fpc);
-	save_fp_regs(p->thread.fp_regs.fprs);
-	p->thread.fp_regs.pad = 0;
-	p->thread.vxrs = NULL;
+	save_fp_ctl(&p->thread.fpu.fpc);
+	save_fp_regs(p->thread.fpu.fprs);
+	p->thread.fpu.pad = 0;
+	p->thread.fpu.vxrs = NULL;
 	/* Set a new TLS ?  */
 	if (clone_flags & CLONE_SETTLS) {
 		unsigned long tls = frame->childregs.gprs[6];
@@ -162,7 +162,7 @@ int copy_thread(unsigned long clone_flags, unsigned long new_stackp,
 
 asmlinkage void execve_tail(void)
 {
-	current->thread.fp_regs.fpc = 0;
+	current->thread.fpu.fpc = 0;
 	asm volatile("sfpc %0" : : "d" (0));
 }
 
diff --git a/arch/s390/kernel/ptrace.c b/arch/s390/kernel/ptrace.c
index d363c9c..52e2e1d 100644
--- a/arch/s390/kernel/ptrace.c
+++ b/arch/s390/kernel/ptrace.c
@@ -59,7 +59,7 @@ void update_cr_regs(struct task_struct *task)
 		if (MACHINE_HAS_VX) {
 			/* Enable/disable of vector extension */
 			cr_new &= ~(1UL << 17);
-			if (task->thread.vxrs)
+			if (task->thread.fpu.vxrs)
 				cr_new |= (1UL << 17);
 		}
 		if (cr_new != cr)
@@ -242,21 +242,21 @@ static unsigned long __peek_user(struct task_struct *child, addr_t addr)
 		/*
 		 * floating point control reg. is in the thread structure
 		 */
-		tmp = child->thread.fp_regs.fpc;
+		tmp = child->thread.fpu.fpc;
 		tmp <<= BITS_PER_LONG - 32;
 
 	} else if (addr < (addr_t) (&dummy->regs.fp_regs + 1)) {
 		/*
-		 * floating point regs. are either in child->thread.fp_regs
-		 * or the child->thread.vxrs array
+		 * floating point regs. are either in child->thread.fpu
+		 * or the child->thread.fpu.vxrs array
 		 */
 		offset = addr - (addr_t) &dummy->regs.fp_regs.fprs;
-		if (child->thread.vxrs)
+		if (is_vx_task(child))
 			tmp = *(addr_t *)
-			       ((addr_t) child->thread.vxrs + 2*offset);
+			       ((addr_t) child->thread.fpu.vxrs + 2*offset);
 		else
 			tmp = *(addr_t *)
-			       ((addr_t) &child->thread.fp_regs.fprs + offset);
+			       ((addr_t) &child->thread.fpu.fprs + offset);
 
 	} else if (addr < (addr_t) (&dummy->regs.per_info + 1)) {
 		/*
@@ -387,20 +387,20 @@ static int __poke_user(struct task_struct *child, addr_t addr, addr_t data)
 		if ((unsigned int) data != 0 ||
 		    test_fp_ctl(data >> (BITS_PER_LONG - 32)))
 			return -EINVAL;
-		child->thread.fp_regs.fpc = data >> (BITS_PER_LONG - 32);
+		child->thread.fpu.fpc = data >> (BITS_PER_LONG - 32);
 
 	} else if (addr < (addr_t) (&dummy->regs.fp_regs + 1)) {
 		/*
-		 * floating point regs. are either in child->thread.fp_regs
-		 * or the child->thread.vxrs array
+		 * floating point regs. are either in child->thread.fpu
+		 * or the child->thread.fpu.vxrs array
 		 */
 		offset = addr - (addr_t) &dummy->regs.fp_regs.fprs;
-		if (child->thread.vxrs)
+		if (is_vx_task(child))
 			*(addr_t *)((addr_t)
-				child->thread.vxrs + 2*offset) = data;
+				child->thread.fpu.vxrs + 2*offset) = data;
 		else
 			*(addr_t *)((addr_t)
-				&child->thread.fp_regs.fprs + offset) = data;
+				&child->thread.fpu.fprs + offset) = data;
 
 	} else if (addr < (addr_t) (&dummy->regs.per_info + 1)) {
 		/*
@@ -621,20 +621,20 @@ static u32 __peek_user_compat(struct task_struct *child, addr_t addr)
 		/*
 		 * floating point control reg. is in the thread structure
 		 */
-		tmp = child->thread.fp_regs.fpc;
+		tmp = child->thread.fpu.fpc;
 
 	} else if (addr < (addr_t) (&dummy32->regs.fp_regs + 1)) {
 		/*
-		 * floating point regs. are either in child->thread.fp_regs
-		 * or the child->thread.vxrs array
+		 * floating point regs. are either in child->thread.fpu
+		 * or the child->thread.fpu.vxrs array
 		 */
 		offset = addr - (addr_t) &dummy32->regs.fp_regs.fprs;
-		if (child->thread.vxrs)
+		if (is_vx_task(child))
 			tmp = *(__u32 *)
-			       ((addr_t) child->thread.vxrs + 2*offset);
+			       ((addr_t) child->thread.fpu.vxrs + 2*offset);
 		else
 			tmp = *(__u32 *)
-			       ((addr_t) &child->thread.fp_regs.fprs + offset);
+			       ((addr_t) &child->thread.fpu.fprs + offset);
 
 	} else if (addr < (addr_t) (&dummy32->regs.per_info + 1)) {
 		/*
@@ -746,20 +746,20 @@ static int __poke_user_compat(struct task_struct *child,
 		 */
 		if (test_fp_ctl(tmp))
 			return -EINVAL;
-		child->thread.fp_regs.fpc = data;
+		child->thread.fpu.fpc = data;
 
 	} else if (addr < (addr_t) (&dummy32->regs.fp_regs + 1)) {
 		/*
-		 * floating point regs. are either in child->thread.fp_regs
-		 * or the child->thread.vxrs array
+		 * floating point regs. are either in child->thread.fpu
+		 * or the child->thread.fpu.vxrs array
 		 */
 		offset = addr - (addr_t) &dummy32->regs.fp_regs.fprs;
-		if (child->thread.vxrs)
+		if (is_vx_task(child))
 			*(__u32 *)((addr_t)
-				child->thread.vxrs + 2*offset) = tmp;
+				child->thread.fpu.vxrs + 2*offset) = tmp;
 		else
 			*(__u32 *)((addr_t)
-				&child->thread.fp_regs.fprs + offset) = tmp;
+				&child->thread.fpu.fprs + offset) = tmp;
 
 	} else if (addr < (addr_t) (&dummy32->regs.per_info + 1)) {
 		/*
@@ -952,18 +952,16 @@ static int s390_fpregs_get(struct task_struct *target,
 			   const struct user_regset *regset, unsigned int pos,
 			   unsigned int count, void *kbuf, void __user *ubuf)
 {
-	if (target == current) {
-		save_fp_ctl(&target->thread.fp_regs.fpc);
-		save_fp_regs(target->thread.fp_regs.fprs);
-	} else if (target->thread.vxrs) {
-		int i;
+	_s390_fp_regs fp_regs;
+
+	if (target == current)
+		save_fpu_regs(&target->thread.fpu);
+
+	fp_regs.fpc = target->thread.fpu.fpc;
+	fpregs_store(&fp_regs, &target->thread.fpu);
 
-		for (i = 0; i < __NUM_VXRS_LOW; i++)
-			target->thread.fp_regs.fprs[i] =
-				*(freg_t *)(target->thread.vxrs + i);
-	}
 	return user_regset_copyout(&pos, &count, &kbuf, &ubuf,
-				   &target->thread.fp_regs, 0, -1);
+				   &fp_regs, 0, -1);
 }
 
 static int s390_fpregs_set(struct task_struct *target,
@@ -972,41 +970,36 @@ static int s390_fpregs_set(struct task_struct *target,
 			   const void __user *ubuf)
 {
 	int rc = 0;
+	freg_t fprs[__NUM_FPRS];
 
-	if (target == current) {
-		save_fp_ctl(&target->thread.fp_regs.fpc);
-		save_fp_regs(target->thread.fp_regs.fprs);
-	}
+	if (target == current)
+		save_fpu_regs(&target->thread.fpu);
 
 	/* If setting FPC, must validate it first. */
 	if (count > 0 && pos < offsetof(s390_fp_regs, fprs)) {
-		u32 ufpc[2] = { target->thread.fp_regs.fpc, 0 };
+		u32 ufpc[2] = { target->thread.fpu.fpc, 0 };
 		rc = user_regset_copyin(&pos, &count, &kbuf, &ubuf, &ufpc,
 					0, offsetof(s390_fp_regs, fprs));
 		if (rc)
 			return rc;
 		if (ufpc[1] != 0 || test_fp_ctl(ufpc[0]))
 			return -EINVAL;
-		target->thread.fp_regs.fpc = ufpc[0];
+		target->thread.fpu.fpc = ufpc[0];
 	}
 
 	if (rc == 0 && count > 0)
 		rc = user_regset_copyin(&pos, &count, &kbuf, &ubuf,
-					target->thread.fp_regs.fprs,
-					offsetof(s390_fp_regs, fprs), -1);
+					fprs, offsetof(s390_fp_regs, fprs), -1);
+	if (rc)
+		return rc;
 
-	if (rc == 0) {
-		if (target == current) {
-			restore_fp_ctl(&target->thread.fp_regs.fpc);
-			restore_fp_regs(target->thread.fp_regs.fprs);
-		} else if (target->thread.vxrs) {
-			int i;
-
-			for (i = 0; i < __NUM_VXRS_LOW; i++)
-				*(freg_t *)(target->thread.vxrs + i) =
-					target->thread.fp_regs.fprs[i];
-		}
-	}
+	if (is_vx_task(target))
+		convert_fp_to_vx(target->thread.fpu.vxrs, fprs);
+	else
+		memcpy(target->thread.fpu.fprs, &fprs, sizeof(fprs));
+
+	if (target == current)
+		restore_fpu_regs(&target->thread.fpu);
 
 	return rc;
 }
@@ -1069,11 +1062,11 @@ static int s390_vxrs_low_get(struct task_struct *target,
 
 	if (!MACHINE_HAS_VX)
 		return -ENODEV;
-	if (target->thread.vxrs) {
+	if (is_vx_task(target)) {
 		if (target == current)
-			save_vx_regs(target->thread.vxrs);
+			save_fpu_regs(&target->thread.fpu);
 		for (i = 0; i < __NUM_VXRS_LOW; i++)
-			vxrs[i] = *((__u64 *)(target->thread.vxrs + i) + 1);
+			vxrs[i] = *((__u64 *)(target->thread.fpu.vxrs + i) + 1);
 	} else
 		memset(vxrs, 0, sizeof(vxrs));
 	return user_regset_copyout(&pos, &count, &kbuf, &ubuf, vxrs, 0, -1);
@@ -1089,19 +1082,19 @@ static int s390_vxrs_low_set(struct task_struct *target,
 
 	if (!MACHINE_HAS_VX)
 		return -ENODEV;
-	if (!target->thread.vxrs) {
+	if (!is_vx_task(target)) {
 		rc = alloc_vector_registers(target);
 		if (rc)
 			return rc;
 	} else if (target == current)
-		save_vx_regs(target->thread.vxrs);
+		save_fpu_regs(&target->thread.fpu);
 
 	rc = user_regset_copyin(&pos, &count, &kbuf, &ubuf, vxrs, 0, -1);
 	if (rc == 0) {
 		for (i = 0; i < __NUM_VXRS_LOW; i++)
-			*((__u64 *)(target->thread.vxrs + i) + 1) = vxrs[i];
+			*((__u64 *)(target->thread.fpu.vxrs + i) + 1) = vxrs[i];
 		if (target == current)
-			restore_vx_regs(target->thread.vxrs);
+			restore_fpu_regs(&target->thread.fpu);
 	}
 
 	return rc;
@@ -1116,10 +1109,10 @@ static int s390_vxrs_high_get(struct task_struct *target,
 
 	if (!MACHINE_HAS_VX)
 		return -ENODEV;
-	if (target->thread.vxrs) {
+	if (is_vx_task(target)) {
 		if (target == current)
-			save_vx_regs(target->thread.vxrs);
-		memcpy(vxrs, target->thread.vxrs + __NUM_VXRS_LOW,
+			save_fpu_regs(&target->thread.fpu);
+		memcpy(vxrs, target->thread.fpu.vxrs + __NUM_VXRS_LOW,
 		       sizeof(vxrs));
 	} else
 		memset(vxrs, 0, sizeof(vxrs));
@@ -1135,17 +1128,17 @@ static int s390_vxrs_high_set(struct task_struct *target,
 
 	if (!MACHINE_HAS_VX)
 		return -ENODEV;
-	if (!target->thread.vxrs) {
+	if (!is_vx_task(target)) {
 		rc = alloc_vector_registers(target);
 		if (rc)
 			return rc;
 	} else if (target == current)
-		save_vx_regs(target->thread.vxrs);
+		save_fpu_regs(&target->thread.fpu);
 
 	rc = user_regset_copyin(&pos, &count, &kbuf, &ubuf,
-				target->thread.vxrs + __NUM_VXRS_LOW, 0, -1);
+				target->thread.fpu.vxrs + __NUM_VXRS_LOW, 0, -1);
 	if (rc == 0 && target == current)
-		restore_vx_regs(target->thread.vxrs);
+		restore_vx_regs(target->thread.fpu.vxrs);
 
 	return rc;
 }
diff --git a/arch/s390/kernel/signal.c b/arch/s390/kernel/signal.c
index 98520c8..49c259c 100644
--- a/arch/s390/kernel/signal.c
+++ b/arch/s390/kernel/signal.c
@@ -105,32 +105,14 @@ struct rt_sigframe
 static void store_sigregs(void)
 {
 	save_access_regs(current->thread.acrs);
-	save_fp_ctl(&current->thread.fp_regs.fpc);
-	if (current->thread.vxrs) {
-		int i;
-
-		save_vx_regs(current->thread.vxrs);
-		for (i = 0; i < __NUM_FPRS; i++)
-			current->thread.fp_regs.fprs[i] =
-				*(freg_t *)(current->thread.vxrs + i);
-	} else
-		save_fp_regs(current->thread.fp_regs.fprs);
+	save_fpu_regs(&current->thread.fpu);
 }
 
 /* Load registers after signal return */
 static void load_sigregs(void)
 {
 	restore_access_regs(current->thread.acrs);
-	restore_fp_ctl(&current->thread.fp_regs.fpc);
-	if (current->thread.vxrs) {
-		int i;
-
-		for (i = 0; i < __NUM_FPRS; i++)
-			*(freg_t *)(current->thread.vxrs + i) =
-				current->thread.fp_regs.fprs[i];
-		restore_vx_regs(current->thread.vxrs);
-	} else
-		restore_fp_regs(current->thread.fp_regs.fprs);
+	restore_fpu_regs(&current->thread.fpu);
 }
 
 /* Returns non-zero on fault. */
@@ -146,8 +128,7 @@ static int save_sigregs(struct pt_regs *regs, _sigregs __user *sregs)
 	memcpy(&user_sregs.regs.gprs, &regs->gprs, sizeof(sregs->regs.gprs));
 	memcpy(&user_sregs.regs.acrs, current->thread.acrs,
 	       sizeof(user_sregs.regs.acrs));
-	memcpy(&user_sregs.fpregs, &current->thread.fp_regs,
-	       sizeof(user_sregs.fpregs));
+	fpregs_store(&user_sregs.fpregs, &current->thread.fpu);
 	if (__copy_to_user(sregs, &user_sregs, sizeof(_sigregs)))
 		return -EFAULT;
 	return 0;
@@ -185,8 +166,7 @@ static int restore_sigregs(struct pt_regs *regs, _sigregs __user *sregs)
 	memcpy(&current->thread.acrs, &user_sregs.regs.acrs,
 	       sizeof(current->thread.acrs));
 
-	memcpy(&current->thread.fp_regs, &user_sregs.fpregs,
-	       sizeof(current->thread.fp_regs));
+	fpregs_load(&user_sregs.fpregs, &current->thread.fpu);
 
 	clear_pt_regs_flag(regs, PIF_SYSCALL); /* No longer in a system call */
 	return 0;
@@ -200,13 +180,13 @@ static int save_sigregs_ext(struct pt_regs *regs,
 	int i;
 
 	/* Save vector registers to signal stack */
-	if (current->thread.vxrs) {
+	if (is_vx_task(current)) {
 		for (i = 0; i < __NUM_VXRS_LOW; i++)
-			vxrs[i] = *((__u64 *)(current->thread.vxrs + i) + 1);
+			vxrs[i] = *((__u64 *)(current->thread.fpu.vxrs + i) + 1);
 		if (__copy_to_user(&sregs_ext->vxrs_low, vxrs,
 				   sizeof(sregs_ext->vxrs_low)) ||
 		    __copy_to_user(&sregs_ext->vxrs_high,
-				   current->thread.vxrs + __NUM_VXRS_LOW,
+				   current->thread.fpu.vxrs + __NUM_VXRS_LOW,
 				   sizeof(sregs_ext->vxrs_high)))
 			return -EFAULT;
 	}
@@ -220,15 +200,15 @@ static int restore_sigregs_ext(struct pt_regs *regs,
 	int i;
 
 	/* Restore vector registers from signal stack */
-	if (current->thread.vxrs) {
+	if (is_vx_task(current)) {
 		if (__copy_from_user(vxrs, &sregs_ext->vxrs_low,
 				     sizeof(sregs_ext->vxrs_low)) ||
-		    __copy_from_user(current->thread.vxrs + __NUM_VXRS_LOW,
+		    __copy_from_user(current->thread.fpu.vxrs + __NUM_VXRS_LOW,
 				     &sregs_ext->vxrs_high,
 				     sizeof(sregs_ext->vxrs_high)))
 			return -EFAULT;
 		for (i = 0; i < __NUM_VXRS_LOW; i++)
-			*((__u64 *)(current->thread.vxrs + i) + 1) = vxrs[i];
+			*((__u64 *)(current->thread.fpu.vxrs + i) + 1) = vxrs[i];
 	}
 	return 0;
 }
@@ -400,7 +380,7 @@ static int setup_rt_frame(struct ksignal *ksig, sigset_t *set,
 	uc_flags = 0;
 	if (MACHINE_HAS_VX) {
 		frame_size += sizeof(_sigregs_ext);
-		if (current->thread.vxrs)
+		if (is_vx_task(current))
 			uc_flags |= UC_VXRS;
 	}
 	frame = get_sigframe(&ksig->ka, regs, frame_size);
diff --git a/arch/s390/kernel/traps.c b/arch/s390/kernel/traps.c
index 7bea81d..97598d1 100644
--- a/arch/s390/kernel/traps.c
+++ b/arch/s390/kernel/traps.c
@@ -19,7 +19,7 @@
 #include <linux/sched.h>
 #include <linux/mm.h>
 #include <linux/slab.h>
-#include <asm/switch_to.h>
+#include <asm/fpu-internal.h>
 #include "entry.h"
 
 int show_unhandled_signals = 1;
@@ -227,7 +227,6 @@ DO_ERROR_INFO(specification_exception, SIGILL, ILL_ILLOPN,
 int alloc_vector_registers(struct task_struct *tsk)
 {
 	__vector128 *vxrs;
-	int i;
 
 	/* Allocate vector register save area. */
 	vxrs = kzalloc(sizeof(__vector128) * __NUM_VXRS,
@@ -236,11 +235,10 @@ int alloc_vector_registers(struct task_struct *tsk)
 		return -ENOMEM;
 	preempt_disable();
 	if (tsk == current)
-		save_fp_regs(tsk->thread.fp_regs.fprs);
+		save_fp_regs(tsk->thread.fpu.fprs);
 	/* Copy the 16 floating point registers */
-	for (i = 0; i < 16; i++)
-		*(freg_t *) &vxrs[i] = tsk->thread.fp_regs.fprs[i];
-	tsk->thread.vxrs = vxrs;
+	convert_fp_to_vx(vxrs, tsk->thread.fpu.fprs);
+	tsk->thread.fpu.vxrs = vxrs;
 	if (tsk == current) {
 		__ctl_set_bit(0, 17);
 		restore_vx_regs(vxrs);
@@ -259,8 +257,8 @@ void vector_exception(struct pt_regs *regs)
 	}
 
 	/* get vector interrupt code from fpc */
-	asm volatile("stfpc %0" : "=Q" (current->thread.fp_regs.fpc));
-	vic = (current->thread.fp_regs.fpc & 0xf00) >> 8;
+	asm volatile("stfpc %0" : "=Q" (current->thread.fpu.fpc));
+	vic = (current->thread.fpu.fpc & 0xf00) >> 8;
 	switch (vic) {
 	case 1: /* invalid vector operation */
 		si_code = FPE_FLTINV;
@@ -297,22 +295,22 @@ void data_exception(struct pt_regs *regs)
 
 	location = get_trap_ip(regs);
 
-	asm volatile("stfpc %0" : "=Q" (current->thread.fp_regs.fpc));
+	asm volatile("stfpc %0" : "=Q" (current->thread.fpu.fpc));
 	/* Check for vector register enablement */
-	if (MACHINE_HAS_VX && !current->thread.vxrs &&
-	    (current->thread.fp_regs.fpc & FPC_DXC_MASK) == 0xfe00) {
+	if (MACHINE_HAS_VX && !is_vx_task(current) &&
+	    (current->thread.fpu.fpc & FPC_DXC_MASK) == 0xfe00) {
 		alloc_vector_registers(current);
 		/* Vector data exception is suppressing, rewind psw. */
 		regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16);
 		clear_pt_regs_flag(regs, PIF_PER_TRAP);
 		return;
 	}
-	if (current->thread.fp_regs.fpc & FPC_DXC_MASK)
+	if (current->thread.fpu.fpc & FPC_DXC_MASK)
 		signal = SIGFPE;
 	else
 		signal = SIGILL;
 	if (signal == SIGFPE)
-		do_fp_trap(regs, current->thread.fp_regs.fpc);
+		do_fp_trap(regs, current->thread.fpu.fpc);
 	else if (signal)
 		do_trap(regs, signal, ILL_ILLOPN, "data exception");
 }